Artefacts

Questions and remarks about code_saturne usage
Forum rules
Please read the forum usage recommendations before posting.
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Artefacts

Post by Asaxena »

Hi

I am trying to solve the large eddy simulation for a vertical cylinder, which has two-inlet boundary conditions. The issue with the simulation is there are some artefacts (possibly due to the mesh) throughout the cylinder in turbulent kinetic energy and velocity calculation. I have tried the simulation with different SGS models for instance wale model, smagorinsky Lilly model, or no model (basically implicit LES). I have used the Least-squares method overextended cell neighbourhood scheme under pressure relaxation factor of 0.5. Moreover, I am using the constant time step option with a reference time step of 5*10-4. We are using liquid metal as a working fluid.

For the meshing, I am using the ICEM software with a general o-grid mesh (structured mesh) methodology.
mesh (1).PNG
setup.log
(27.94 KiB) Downloaded 62 times
I have attached the setup log file. please have a look.
Attachments
artefacts.PNG
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

The artefacts closest to the center of the mesh really seem to correlate with the mesh (you can visualize mesh quality criteria to compare).

For the others, it seems less clear.

A relaxation of 0.5 for pressure seems quite a lot, and I am not sure it is recommended for LES (you probably lose the order 2 in time using relaxation).

I would recommend :

- Reduce the relaxation for pressure.

- Add some sweeps for resolution if you are using less than the defaults(or LES, 10 for velocity and 5 for pressure, if I remember correctly).

- Possibly add 2 to 5% of upwind to the velocity resolution

- Then possibly switch back to Green-Gauss gradient (the hybrid one with least-squares face values estimation should be good, keeping the extended neighborhood if needed for that).

Also, though I am not a specialist, is your inlet turbulence level sufficient ? Do you use the synthetic eddy method option for inlet turbulence ? If not, and your domain is not long enough, perhaps this may be related too (I would prefer an LES user confirm this).

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi,

Thank you for your reply.
As per your suggestion I have checked the mesh quality criteria and some non_ortho mesh has been detected (please find the attached image). Additionally I have tried the simulation with a blending factor of 0.95 for velocity (second order linear upwind) and with pressure relaxation value  of 0.2. I have tried the simulation with both numerical schemes  the green-gauss with iterative scheme of non-orthogonality as well as green-gauss with least square face values. In both of the cases with the above mentioned conditions (pressure relaxation = 0.2; blending factor = 0.95) the simulations have diverged. The simulation have diverged in a very early stage (i.e. after 2 or 3 iterations). 
non-orthogonal.JPG
(17.39 KiB) Not downloaded yet
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

Does the computation run OK with another turbulence model (k-epsilon, k-omega, or Rij) ? If so, it could be interesting to try starting it that way and restarting in LES.

Otherwise, you could start in LES with your initial options and then restart with the recommended ones. I am not sure this will be enough, but it may help.

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi
Thank you for your reply,
The simulation has been converged for the LES simulation (restart with RANS simulation) or worked well in other words if we used the code-saturne V6.0.0. I have used the least square gradient scheme. But we noticed that there are mesh artifacts in the results. However, the simulation has diverged in the case of the RANS simulation when we used the code-saturne V7.0.0 with the green-gauss-based iterative scheme, pressure relaxation =0.3, blending factor for velocity 0.95.

Let me make it more clear through two cases which are described below:

case-1 LES simulation (restart with RANS), pressure relaxation =0.5, least square method overextended cell neighborhood; Solution has been converged; Issue: mesh artifacts in the solution

case-2 RANS simulation (diverged), pressure relaxation =0.3, green gauss with iterative handling of non-orthogonality.
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

It may be interesting to test a middle-ground solution, using the hybrid Gauss-Least-squares gradient (LSQ for face value prediction instead of an iterative process, the Gauss for the conservative formulation). This allows using an extended neighborhood for the LSQ part, though in most cases, the final behavior is more similar to that with the iterative algorithm than with pure least-squares.

If the artefacts are present in the least-quares version, which presents the smoothest solution, I double they will disappear with another option though, unless other subtle (consistency, local scheme order) aspects kick in, which is possible.
I would guess more sweeps would be helpful here, but in LES ,you already have 5 sweeps for the pressure. You can check if going to 10 might help. Setting the "iswdyn" option (check in cs_user_parameters-base.c example) might also help in combination with this.

And since LES is quite sensitive to mesh quality, and you will inherently have more numerical viscosity in tetrahedral/prism zones, working around this might not be easy. Some colleagues have success running LES on tetrahedral meshes, but I am not sure which specific numerical options they adapt (compared to defaults) for this.

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi,

So, as per your suggestion, I have tried the green-gauss with least-squares gradient face values gradient reconstruction type with automatic extended cell neighbors option. I have increased the sweep value from 5 to 10 for pressure. However, the simulation has diverged at a very early stage after 2 iterations with the following messag (symmetric-gauss seidel: error divergence solving for velocity).

Thanks
Ashish Saxena
Yvan Fournier
Posts: 4070
Joined: Mon Feb 20, 2012 3:25 pm

Re: Artefacts

Post by Yvan Fournier »

Hello,

I'll need to check with colleagues who ran some LES on tetrahedral meshes for more suggestions/details. I know in one case they needed to do specific filtering (I don't know if it was post-processing or upstream or both), so their solution might filter their artefacts. I have not had the time to look into more detail in what they did, so I'll need their help here. I'll ask them if they can take a look at this thread.

Best regards,

Yvan
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi,

Thank you for your reply,
please do the needful soon.



Regards
Ashish Saxena
Asaxena
Posts: 16
Joined: Thu Mar 17, 2022 12:44 pm

Re: Artefacts

Post by Asaxena »

Hi Yvan
Could you Please discussed/involved the guy in the thread of discussion you talking about.
Post Reply